Policy

US Government Secures Early Access to Frontier AI Models from Microsoft, Google, and xAI for Security Testing

On May 5, 2026, the Center for AI Standards and Innovation (CAISI) at the US Department of Commerce announced agreements with Microsoft, Google DeepMind, and xAI to provide early access to new frontier AI models for national security and security risk evaluations before deployment, expanding prior partnerships and responding to concerns over advanced capabilities like those demonstrated by Anthropic's Mythos.

Tech Insights Reporter May 5, 2026 7 min read Washington, D.C.

TLDR

The US government's CAISI announced on May 5, 2026, new agreements with Microsoft, Google (DeepMind), and xAI for pre-deployment access to their frontier AI models. The center will evaluate the models for national security risks, including cybersecurity threats, and conduct collaborative research on capabilities and safeguards. This builds on existing deals with OpenAI and Anthropic and comes amid growing alarm over models like Anthropic's Mythos, which excels at identifying vulnerabilities and generating exploits.

CAISI Agreements and Scope

CAISI, housed within the National Institute of Standards and Technology (NIST) at the Department of Commerce, serves as the primary hub for AI model testing and evaluation. Under the new agreements:

Developers will provide early access to new models, often with safety guardrails removed for probing.
CAISI will assess capabilities, security risks, and national security implications before public deployment.
The program has already completed more than 40 evaluations of cutting-edge models.

The agreements fulfill pledges from the Trump administration's July 2025 AI Action Plan to partner with tech companies on vetting for national security risks.

Company-Specific Commitments

Microsoft: Will collaborate with US government scientists on adversarial assessments that probe unexpected behaviors, misuse pathways, and failure modes. Work includes co-developing systematic, reproducible evaluation approaches, shared frameworks, datasets, and workflows for safety, security, and robustness. Microsoft also signed a parallel agreement with the UK's AI Security Institute (AISI) for frontier safety research and societal resilience studies.
Google DeepMind and xAI: Agreed to provide models for similar evaluations focused on capabilities and risks.

CAISI Director Chris Fall emphasized that "independent, rigorous measurement science is essential to understanding frontier AI and its national security implications."

Context: Rising Concerns Over Advanced Capabilities

The moves reflect heightened US government focus on dual-use risks from frontier models. Recent capabilities in models like Anthropic's Mythos—particularly its ability to autonomously find and chain vulnerabilities at scale—have raised alarms about AI supercharging cyberattacks and other threats.

Developers frequently share models stripped of safeguards so evaluators can identify potential misuse vectors.

This builds on 2024-era agreements (originally under the US AI Safety Institute) with OpenAI and Anthropic, now expanded under the current administration.

Why this story matters

These agreements institutionalize government-industry collaboration on AI evaluation at the frontier level, providing structured insight into model capabilities before widespread deployment. They signal a maturing approach to balancing innovation with national security, focusing on concrete risks like AI-enabled hacking while fostering shared evaluation methodologies. As more labs join, it could set de facto standards for pre-release scrutiny and influence global practices.

Sources

Reuters: “Microsoft, Google and xAI to give US government early access to AI models for security checks” (Courtney Rozen, Aditya Soni; published May 5, 2026). https://www.reuters.com/legal/litigation/microsoft-xai-google-will-share-ai-models-with-us-govt-security-reviews-2026-05-05/
Microsoft On the Issues: “Advancing AI evaluation with the Center for AI Standards (US) and Innovation and the AI Security Institute (UK)” (Natasha Crampton; published May 5, 2026). https://blogs.microsoft.com/on-the-issues/2026/05/05/advancing-ai-evaluation-with-the-center-for-ai-standards-us-and-innovation-and-the-ai-security-institute-uk/
CAISI statements via NIST/Commerce (cross-referenced in reporting).
Background on prior agreements and Mythos capabilities from contemporaneous coverage.

Featured Image Alt Text

Conceptual image of US government and tech companies collaborating on AI model evaluation, with symbolic elements of security testing, national flags, and frontier model interfaces